Ontology-Based Data Integration from Heterogeneous …web.mit.edu/cron/project/CUPUM2015/proceedings/Content/... · ... age and gender distribution ... Linked Data [3] and Semantic

Ontology-Based Data Integration from

Heterogeneous Urban Systems: A Knowledge

Representation Framework for Smart Cities

Achilleas Psyllidis

Abstract

This paper presents a novel knowledge representation framework for smart

city planning and management that enables the semantic integration of

heterogeneous urban data from diverse sources. Currently, the combination of

information across city agencies is cumbersome, as the increasingly available

datasets are stored in disparate data silos, using different models and schemas

for their description. To overcome this interoperability barrier, the presented

framework employs a modular and scalable system architecture, comprising a

comprehensive ontology capable of integrating data from various sectors

within a city, a web ontology browser, and a web-based knowledge graph for

online data discovery, mapping, and sharing across stakeholders. Linked Data,

Semantic Web technologies, and ontology matching techniques are key to the

framework’s implementation. The paper ultimately showcases an application

example, where the framework is used as a semantic enrichment mechanism

in a platform for urban analytics, focusing particularly on human-generated

data integration.

_________________________________________________________ A. Psyllidis

Chair of Hyperbody – Digitally-driven Architecture, Department of

Architectural Engineering & Technology,

Faculty of Architecture and the Built Environment,

Delft University of Technology (TU Delft), 2628 BL, Delft, The Netherlands

Email: [email protected]

CUPUM 2015 240-Paper

1 Introduction

At present, large-scale urban data are produced from diverse sources and

increasingly become publicly available from governmental authorities. Each

source reflects a specific aspect of the urban environment and the generated

datasets vary in scale, speed, quality, format, and, most importantly,

semantics. In principle, municipal data are mostly static or semi-static, in the

form of demographic records, such as census statistics on population, age and

gender distribution, average income, and land uses. Besides static records,

planning stakeholders gradually utilize dynamic streams of information

stemming from sensor resources, in relation to transportation flows, weather

status, and environmental conditions. However, in the current context, city-

related organizations create and operate on datasets based on each sector’s

particular purposes and problems at hand. Any correlation of information

across different departments is presently performed in a manual fashion,

hence requiring great amounts of time and effort.

This data interoperability barrier is further strengthened by the use of

different data models and schemas across city agencies [13]. Equally, in urban

planning and decision-making procedures each urban system is often

approached separately, as a result of the complexity and diversity among the

city data silos. This leads to an emerging need for frameworks that allow for

interoperable urban data exchange and reuse, by systematically harnessing the

combined potential of diverse data sources. Such an approach is particularly

important within a smart city framework.

Motivated by this challenge, the paper introduces a novel knowledge

representation (KR) framework for planning smart cities that enables the

semantic integration of heterogeneous urban data stemming from diverse

sources. The framework consists of three main components, namely (a) an

ontology that formally describes the different city sectors, urban systems, the

respective data sources, and defines the relations among them, (b) a web

ontology browser that provides full access to the aforementioned ontology

online, and (c) an interactive knowledge graph to facilitate the exchange of

shared semantic definitions and correlations among multiple stakeholders,

through a web-based graphical user interface (GUI).

The ontology for smart city planning and management presented in this

paper is currently the first one in this domain that defines concepts and named

relationships, linking both the different urban systems together and the data

generated within them. By additionally implementing alignments with

CUPUM 2015 Psyllidis 240-2

multiple external ontologies and controlled vocabularies, it can be used to

semantically annotate data from heterogeneous city sectors in a machine-

processable way. Moreover, the developed web ontology browser provides

developers and other stakeholders with online access and navigation

possibilities through the complete ontology hierarchy. By encompassing

Linked Data [3] and Semantic Web principles [1, 7], it allows concept

discovery, browsing over named relations, and links to aligned concepts and

properties. Similarly, the interactive knowledge graph constitutes a novel

visualization system for exploring the various entity clusters and relations

captured by the ontology. It allows city stakeholders to perform term

searching, concept clustering, and relations’ discovery through an intuitive

web-based platform.

The aim of the introduced KR framework – coined OSMoSys – is to assist

in the mutual interaction among urban planners, city managers, decision-

makers, and other city stakeholders by providing tools that are capable of

operating across the diverse data silos. It also aims to simplify the processes

of data mapping, sharing, and reuse across agencies. The contributions of this

work are mainly three.

1. A novel ontology for smart cities that enables data integration from

heterogeneous urban sectors and sources, irrespective of the different

qualities, speeds, and scales. An additional innovative aspect of this

ontology refers to the integration of concepts for human-generated data,

coming from smart devices and social media platforms, being increasingly

important sources of knowledge for the domains of urban analytics and

decision-making.

2. A web ontology browser and an interactive knowledge graph that

complement the processes of concept discovery and data mapping, by

respectively providing an easy-to-use online ontology navigation platform

with links to external data models and a web-based visualization system

for intuitive concept exploration1.

3. An application example of using the KR framework as a semantic

enrichment mechanism in a recently developed system that supports the

analysis and integration of heterogeneous urban data.

1 The web-based knowledge graph and ontology browser are accessible via the following link:

http://osmosys.hyperbody.nl

CUPUM 2015 Ontology-Based Data Integration from Heterogeneous Urban... 240-3

The rest of the paper is structured as follows. In Section 2 the related work

is described. Section 3 introduces and presents the proposed method. Section

4 outlines the architecture of the KR framework and describes its three main

constituent components. In Section 5 a use example of the knowledge

framework is showcased. Ultimately, Section 6 summarizes the conclusions

and discusses future lines of research.

2 Related Work

In the wider domain of computational urban planning there have recently been

several attempts to tackle the complexity of contemporary city systems and

the increasing amount of data they generate, through the use of ontology-

based models [15]. However, in most of these cases, the developed ontologies

either cater for an individual urban system or represent specific attributes of

the urban environment. To this end, creating new ontology models, without

establishing links among them, generally poses a risk as to merely adding up

to the multiplicity of already existing, yet discordant, data models and

schemas. The examples described in the following paragraphs showcase

relevant approaches that serve as stepping-stones on the development of the

broader framework proposed in this work.

To sufficiently describe urban space and facilitate interoperability among

urban design applications, Montenegro et al. [14] developed an ontology for

land use planning. Having as its starting point the Land Base Classification

Standards (LBCS) by the American Planning Association, the ontology

provides semantically annotated land use descriptions of spatial data. The

main limitations of this study relate to the adherence of the developed

ontology to a single model and, subsequently, to a particular aspect of the

urban environment. However, the complexity of urban space requires links to

multiple external ontologies or conceptual schemas in order to be

comprehensively described and understood.

Métral et al. [13] describe the benefits of following an ontology-based

approach for providing semantically enriched 3D city models. Besides

presenting an ontology of the CityGML model [11], which incorporates both

geometrical and topological concepts of urban objects, the paper discusses the

Ontology of Urban Planning Processes (OUPP), with a focus on soft mobility.

The work presents similar limitations to the previous one, yet it indicates an

alignment with concepts from an external model, namely the Ontology of

Transportation Networks (OTN). A further collection of ontologies,


specifically created for projects in the domains of urban planning and

development, is presented in [10]. In this case, the described knowledge

models are accompanied by real-world case studies, covering an array of

aspects that range from urban mobility to building regulations.

Closer to the intentions of this paper is the work of Bellini et al. [2]. Their

paper describes the construction process of a knowledge base for smart-city

services. An interesting element of this work is the introduction of an

ontology for smart-city services that allows the integration of heterogeneous

data from multiple sources, supported by links to external vocabularies.

Nevertheless, it is mainly focused again on a single urban system (i.e.

transport) and thus faces similar limitations to those of the aforementioned

examples.

Unlike previous approaches, the KR framework for smart cities presented

in this paper offers a unique framework that combines an innovative ontology,

allowing data mapping from different urban systems, with web-based tools for

intuitive concept sharing among city stakeholders.

3 Method

To address the challenges posed by cumbersome and disparate data silos

across city agencies, the demonstrated KR framework follows an ontology-

based approach to integrating urban data from different sectors. To this end, a

comprehensive ontology for smart cities was developed that formally

describes the different urban systems (e.g. energy, waste, water, transport,

buildings etc.); the respective data sources (e.g. spatial statistics, sensor data,

social data etc.); the smart city technology enablers (e.g. interoperability

types, connectivity, computational resources etc.); and defines the relations

among them. It also represents the various urban sectors/agencies and

includes broader concepts that allow data mapping from different

departments. The intention is to enable the exploitation of the reasoning

potential inherent to ontologies, as well as the data discovery possibilities

through expressive querying.

However, when combining data of different speeds (i.e. static, semi-static,

streaming), scales, and qualities, stakeholders are generally confronted with

syntactic, schematic, and semantic heterogeneities. These respectively refer to

differences in data syntax, database schemas, and meaning interpretation.

Notwithstanding, emphasis in this case is put on semantics. Providing shared


semantic definitions is essential for both data integration and reuse into

different contexts. Especially when it comes to sharing information across

agencies that might refer to the same component, yet originally using a

different term or data model to describe it. Therefore, entities of the developed

ontology for smart cities are aligned with concepts and properties from

multiple external ontologies and vocabularies, to ensure the interrelation

among city sectors and further facilitate the data integration process. To

achieve this, ontology matching techniques [9] are used. As the proposed

semantic model aims to provide an overarching framework that encompasses

the various urban systems, it is crucial for it to enable links among the already

existing knowledge resources, which cater for more specific aspects of the

urban environment. In the previous section it was shown that single sector-

specific ontologies might well form such knowledge resources, yet

insufficient in addressing cross-sector concepts. Mapping to a global

framework is thus paramount, and the proposed one in this paper is capable of

providing it.

The above-described domain ontology constitutes one of the three

components the OSMoSys KR framework introduces, as part of its data

integration approach. To further facilitate the collaboration among the

different city stakeholders, it additionally incorporates a web ontology

browser and a web-based knowledge graph for smart cities. The role of the

two latter components is to provide online, intuitive, and interactive user

interfaces for exploring the underpinning ontology. In this way, multiple

stakeholders of different backgrounds, such as urban planners, decision-

makers and other municipal authorities, possessing various levels of expertise,

can equally access and discover concepts and relations represented in the

semantic model. The web ontology browser incorporates the complete

taxonomy of ontology concepts (classes, object and data properties,

individuals etc.) and offers multiple exploration possibilities online to people

who are not trained in knowledge representation formalisms, without

requiring dedicated software platforms to be downloaded. By adopting

Semantic Web and Linked Data principles, it provides direct links to other

knowledge repositories that cater for more specific purposes, thus

encouraging concept and data model reuse. Finally, the web-based knowledge

graph offers an interactive GUI for visual exploration, navigation through

linked open data, and concept discovery. It visually represents the semantic

model as a dynamic graph of concepts and relations, using state-of-the-art

web and data visualization technologies.


4 OSMoSys: The Knowledge Representation Framework

The OSMoSys KR framework for smart city planning and management adopts

a modular architecture, so as to address the several scalability challenges

posed by the diversity of urban data, as well as by the complexity of cities and

respective agencies. These modules namely refer to (a) the Ontology for smart

city planning and management, (b) the Web Ontology Browser, and (c) the

web-based Knowledge Graph. Each of the aforementioned components has a

specific functionality and serves a particular purpose. As such, they can be

deployed independently, yet all three are simultaneously linked to each other.

Owing to this modular architecture, the KR framework can be extended,

accommodating new concepts, properties and data sources, and can further be

shared and reused across sectors with relatively low effort. When a new entity

is introduced to the system, all components are automatically updated.

Thereby, the framework is capable of serving the changing needs of city

sectors, based on the issue at hand, while adapting to the dynamic nature of

urban systems. This section provides a detailed description of the modules

that compose the OSMoSys framework.

4.1 Ontology for Smart City Planning and Management

4.1.1 Domain and Scope Definition

In order for the ontology to constitute an encompassing framework for the

domains of smart city planning and management, its taxonomy is organized

around eleven principal classes (namely, Case, Decision, Document, Event,

Method, Place, Sampling, Situation, Technologies, TemporalEntity,

WebEntity) that represent broader and cross-sector domain concepts. The

latter are further specialized, by including subclasses that enable the

representation of the various urban systems, their respective data sources, the

different agent types, the diverse city services, and the technologies that

facilitate their interconnection. In this regard, the complete ontology

comprises 121 classes, 82 object, data, and annotation properties, as well as

23 individuals and data types, representing a total of 226 entities (based on

various standards, relevant external ontologies, and controlled vocabularies,

as described in Sect. 4.1.2).

The developed ontology defines a common semantic model describing

data from disparate sectors within a city in a machine-processable way, with a


focus on planning and management. To this end, it constitutes the first

approach in this domain. Its scope is to specifically facilitate the mapping of

heterogeneous data models, already in use in cities, to its overarching

semantic framework. Nevertheless, it does not intend to replace any of the

various existing data models, but rather to enable the interoperable

information exchange among them and across city agencies.

4.1.2 Reuse of Existing Semantic Models

The ontology reuses terms and concepts from various European and American

standards and roadmaps on smart cities, as well as from relevant external

ontologies and controlled vocabularies. The former namely refer to the

Publicly Available Specifications (PAS) on Smart Cities [4] and the Smart

City Concept Model [5], the Smart Cities Readiness Guide [17], and the

Operational Implementation Plans of the European Innovation Partnership on

Smart Cities and Communities [8].

Besides, terms from the following vocabularies are reused: dc: and dct: a

collection of metadata terms maintained by the Dublin Core Metadata

Initiative; foaf: a dictionary of named properties and classes, integrating

social, information, and representational networks; gml: an XML-based

grammar and encoding standard of the Open Geospatial Consortium (OGC)

for describing geographical features; schema: a set of schemas for HTML

markup and structured data interoperability on the web for search

optimization; skos: a representation standard for knowledge organization

systems; vann: a controlled vocabulary for semantic annotations. Classes and

properties of the proposed semantic model are aligned with the following

external ontologies: DUL (Dolce+DnS Ultralite upper level ontology);

CityGML (unified model for 3D city representations); dbpedia-owl; OTN

(Ontology of Transportation Networks); SSN (Semantic Sensor Network

ontology); OWL-Time (ontology of temporal concepts). Finally, the ontology

fully complies with the following data modeling vocabularies: owl (Web

Ontology Language); owl2xml; rdf (Resource Description Framework); rdfs

(RDF Schema); xsd (XML Schema).


4.1.3 Ontology Development and Taxonomy Definition

The OWL2 (Web Ontology Language)2 specification was used for developing

the ontology for smart cities, and it was further built with the Protégé 4.3.0

ontology development platform3. Mapping the various standards mentioned in

the previous paragraph into the demonstrated ontology, gives stakeholders the

opportunity to take advantage of its inference and reasoning potential. In

addition, it facilitates the reusability and extensibility of the standards

themselves, especially through the web-based components of the OSMoSys

model (the Web Ontology Browser and the Knowledge Graph) that

complement the described ontology.

Moreover, the variety of vocabularies and external data models

incorporated in the presented ontology indicates its intention to cover the

complexity and diversity of city-related information, ranging from spatial and

geometric features (e.g. gml, CityGML, OTN) to sensor, social, and

information networks (e.g. SSN, foaf), and time intervals (OWL-Time). Yet, it

substantially extends them by introducing multiple new concepts and

relations, so as to also address the aforementioned smart city roadmaps and

standards. A representative sample of these extensions is further described in

the following paragraphs (for clarity purposes, concepts of reused

vocabularies or data models are indicated by the accompanying namespace

prefix).

One example case is that of the schema:Place principal class. The latter is

further specialized, including concepts such as schema:AdministrativeArea,

dbpedia-owl:City, Item, foaf:Agent, DUL:InformationObject,

dct:PhysicalObject, Service, and UrbanSystem. This taxonomy development

approach allows concepts, referring to both physical and virtual features of the

city, to inherit (rdfs:subClassOf) the spatial properties (geo-reference) of the

top-level class schema:Place. Subsequently, the data mapped to these

concepts, irrespective of the city sector they refer to, will immediately inherit

these properties as well. This is highly essential for the contemporary

planning procedure that not only engages with the physical elements of cities

and urban systems but increasingly also with the digital information

embedded in them. The UrbanSystem class, introduced by the presented

2 http://www.w3.org/TR/owl2-profiles/

3 http://protege.stanford.edu


ontology, enlists a hierarchy of subclasses referring to the diverse urban

systems (e.g. BuiltEnvironment, Energy, Environment, Healthcare, Transport,

Waste etc.), while the Service class models the corresponding services

provided by the different sectors. The respective data produced across the

various city systems (UrbanSystem) are modeled as instances of the

UrbanSystemsData class, which is a sub-concept (rdfs:subClassOf) of the

DUL:InformationObject, and can further be used (hasRoleIn) in decision-

making (Decision class). Hence, through the previously described taxonomy

of classes and relations (object properties), the ontology enables the

interoperability among various city sectors, its services, and its corresponding

data sources.

With respect to the increasing amount of sensor devices in contemporary

urban environments, the ontology incorporates the ssn:Sensor concept (under

dct:PhysicalObject) of the Semantic Sensor Network (SSN) Ontology [6], yet

it specializes and extends it, so as to better describe the present variety of

sensor resources. Thus, the SensorResource concept is introduced that

incorporates both the ssn:Sensor and the SensorSystem classes (e.g. a weather

station comprising multiple sensors in one system, is not sufficiently

described by the ssn:Sensor concept). Sensor resources in the demonstrated

ontology are part of (dct:isPartOf) a wider sensor network (SensorNetwork)

or a sensor web (SensorWeb under the WebEntity principal class).

One of the most innovative aspects of the ontology concerns its ability to

integrate entities referring to human-generated data in cities. Stemming

mainly from smart devices and social media platforms, they constitute

increasingly important sources of information, especially for the domains of

urban analytics and decision-making. For this, the ontology introduces

multiple concepts to enable their modeling. For instance, as a specification of

the ssn:Sensor class, the HumanSensor entity is added. The latter is defined as

a person who reacts to stimuli and/or events and occasionally makes

observations about the physical world. Therefore, it is made equivalent

(owl:equivalentClass) to foaf:Person, for the system to realize that instances

belonging to this entity solely refer to humans, instead of devices (e.g.

ssn:Sensor or SensorSystem). As such, it is also related to the broader

concepts of foaf:Agent, foaf:Group, DUL:Community, and foaf:Organization.

Besides the SocialData and SocialWebData classes are introduced, under

DUL:InformationObject, to respectively model urban data from census

records or demographics, and streams from social media platforms (e.g.

Twitter, Instagram, Foursquare etc.), which differ from sensor streams


(SensorData). These concepts are further related – through OWL object

properties – to urban systems for further integration in the planning process.

The ontology also includes a Crowdsourcing class, to describe a

contemporary form of urban sensing based on human input.

Time intervals and diversities in data speed are also taken into account,

modeled through classes such as time:TemporalEntity, time:Interval,

time:Instant, StaticSampling, Semi-StaticStampling, and DynamicSampling.

4.1.4 Ontology Matching

The alignment of the developed ontology with the multiple controlled

vocabularies and external semantic models is achieved through the use of

ontology matching techniques. The latter enable the discovery of mutual

relations and correspondences among entities belonging to different

ontologies. In this case, correspondences are established through equivalence

relations (owl:equivalentClass), links between individuals (owl:sameAs), and

structural equivalence indications between entity IRIs (Internationalized

Resource Identifiers) with the use of IRI refactoring. All the previous

matching measures use basic similarity, meaning that they make comparisons

based on a particular characteristic of the aligned entities (e.g. name,

dct:description, IRI string etc.) [9].

4.1.5 Inconsistency Check and Evaluation (Reasoning)

The developed ontology has fully been evaluated for potential anomalies and

inconsistencies, by performing tests with multiple open-source reasoners,

namely FaCT++4, HermiT 1.3.85, and Pellet6. The inferred ontology class

hierarchy proved to be fully consistent in all aforementioned tests.

4 http://owl.man.ac.uk/factplusplus/

5 http://hermit-reasoner.com

6 http://clarkparsia.com/pellet


4.2 Web Ontology Browser

To provide city stakeholders with straightforward access to the above-

described ontology, OSMoSys further incorporates a Web Ontology Browser

(WOB). The latter offers a web-based user interface (UI) for navigating

through the complete semantic model hierarchy, without requiring

computational skills in ontology development environments or relevant

software. Unlike the majority of ontologies and data models, which remain

accessible to a limited group of experts, the WOB allows open access to any

interested party. Thereby, it aims to provide a tool for bridging together the

disparate urban data silos, thus facilitating the collaboration among different

city agencies.

The WOB for smart city planning and management is fully indexed,

containing the complete list of entities (i.e. classes, object and data properties,

annotations, individuals, and data types) of the developed ontology. Besides,

any classes or properties mapped to entities of related ontologies or controlled

vocabularies, include links to the external IRIs, in accordance with Linked

Data and Semantic Web principles (based on owl:sameAs and rdfs:seeAlso

statements).

The UI layout of the WOB is organized in three panes, providing different

browsing possibilities and views of the ontology (Fig. 1). The upper-left pane

lists the different ontology entities in groups of classes, object, data, and

annotation properties, individuals, and data types. It also contains an option

for returning back to the general overview. When any of the previous entity

groups is selected, a complete index of the corresponding concepts in

alphabetical order appears on the lower-left pane of the UI. Finally, the main

pane accommodates the full semantics, descriptions, and annotations of each

selected entity, as well as its relations to other classes, and links to external

aligned concepts. Definitions and other annotations, object properties and

taxonomic hierarchy information, as well as description logics formalisms are

clearly organized in different blocks within the main pane, so that they are

easily distinguishable. The user can interactively browse the different entities

and explore relations, either through the side-pane indexes or by directly

clicking on any term included in the main pane. In addition, when a new

concept is introduced to the system, the WOB is automatically updated, hence

facilitating the data mapping process.


Fig. 1. User interface of the Web Ontology Browser

4.3 Web-based Knowledge Graph

To further facilitate the data sharing and concept discovery processes across

heterogeneous sectors, the OSMoSys framework also introduces a novel web-

based Knowledge Graph (KG), to allow visual and interactive exploration of

the underlying domain ontology, as well as linked urban data mapped into it

(Fig. 2). Alongside the WOB module, it further provides a dynamic, graph-

based representation of the cross-sector entities and their relations. As

previously discussed, each of the OSMoSys components can be used

independently, owing to its modular architecture, yet the combination of all

three exploits its full potential. In particular, the KG mainly focuses on visual

exploration and discovery of concepts and interrelations, whereas the WOB

primarily facilitates the data mapping and integration process. Formal

description, semantic annotation and introduction of new city-related entities

take place in the backend ontology module, which in turn supplies and

informs the other two components of OSMoSys.


Fig. 2. User interface of the web-based Knowledge Graph. The dynamic graph layout

displays ontology classes as nodes, and relations as edges. When hovering over a

node, its label is highlighted. Different levels of detail are provided through semantic

zooming

The KG receives as input urban data from diverse sources in RDF

(Resource Description Framework) format, mapped into the above-described

ontology, and displays them on a force-directed graph layout. An innovative

aspect of the presented KG lies in its ability to visualize heterogeneous, yet

semantically annotated, urban data as instances of ontology classes along with

their relations in a single web-based platform. In this way, it aims to enable

the interoperability among the various urban systems and, thereby, support

planners and decision-makers in simultaneously correlating information from

different sectors. Unlike the current manual and labor-intensive connections

across data sources, the developed GUI provides stakeholders with easy

navigation through the multiple domain concepts and allows them to

effortlessly discover relations among entities and data.

The graph layout displays ontology classes as nodes, and relations (object,

data, and annotation properties) as edges. Node size variations indicate

whether an entity refers to a principal class or a subclass. The default top-level

class, hence the largest node, is owl:Thing, since every individual, according

to the OWL specification, is an instance of this particular class. In full display

mode, besides owl:Thing, the user can easily recognize the eleven principal

classes, around which the semantic model is organized (described in section

4.1). As a user zooms in, the graph displays different levels of detail,

gradually making all subclass labels visible, in a process called semantic

zooming. Hovering over a class node highlights its label along with its


immediate relations to other classes, for visual clarity purposes. By further

selecting a highlighted graph entity, an information pane pops up on the right,

containing additional details about the class itself, as well as an index of all

related entities. Moreover, the main graph display shows an isolated view of

the selected concept and its connections. The user can easily return to the full

network view, by choosing the corresponding option provided by the GUI.

Navigation is also possible through the information pane index.

One of the key features of the KG is the search module. Thereby, the user

is provided with the possibility to discover a concept, by performing keyword

search, without necessarily knowing the exact class name in advance. All

relevant results will appear as a list below the search field, on the left floating

menu. In this way, the user can navigate through the various results and,

correspondingly, highlight the selected concepts and their relations on the

main display. Besides the search modules, users are also provided with a

group selection tool. The latter is currently capable of grouping concepts of

the semantic model, based on their function as top-level, principal, or sub-

classes.

The web-based KG was created with the use of the Sigma.js JavaScript

library7. The force-directed graph layout of the underlying ontology was built

with Gephi8, in combination with the Yifan Hu multilevel algorithm [12]. Pre-

processing of the ontology graph was implemented using OntoGraf9 and

Graphviz10. RDF data are stored in and retrieved from a SparkleDB

triplestore11. The KG is automatically updated, based on changes made in the

underlying semantic model.

7 http://sigmajs.org

8 https://gephi.github.io

9 http://protegewiki.stanford.edu/wiki/OntoGraf

10 http://www.graphviz.org

11 http://www.sparkledb.net


5 Use Case

OSMoSys has successfully been used as a back-end semantic enrichment and

data integration mechanism in a recently developed web-based system12 that

supports the analysis, integration, and visualization of large-scale and

heterogeneous urban data. An interesting case for urban decision-making

refers to city-scale events, as they can have a strong impact on a variety of

urban aspects, such as mobility flows and occupancy levels in specific public

spaces in the city. Different stakeholders (e.g. event organizers, municipal

authorities etc.) can benefit from real-time (or near real-time) urban analytics,

in relation to the events, for making better-informed decisions. Yet, this is a

challenging task, as it requires the simultaneous combination and evaluation

of different data sources. In this context, OSMoSys enabled the semantic

annotation and integration of heterogeneous data ingested in the

aforementioned system, with application to the monitoring and assessment of

a) the Amsterdam Light Festival (ALF) 2015, and b) Milano Design Week

(MDW) 2014.

The duration of ALF was from November 27, 2014 to January 18, 2015.

MDW concerned a shorter time period, namely from April 4, 2014 to April

14, 2014. Apart from multiple (semi-)static datasets (e.g. municipal

demographic records, event maps with installations etc.), the system analyzed

data streams from traffic sensors and social media platforms, namely Twitter

and Instagram. For ALF, a total of 26.740.669 geo-referenced tweets and

15.959.566 Instagram posts were collected. In the MDW case, the system

collected 1.879.187 geo-referenced tweets and 1.090.237 Instagram posts.

These data streams referred to posts in the proximity of the installations or

event area.

The ingested data sets and streams were mapped into the developed

ontology as instances of classes to facilitate the processes of data exploration,

analysis, querying, and visualization (Fig. 3). In both example cases, the

collected data were mapped as instances of the class dct:Event, which

connects them to a specific site (through the object property atPlace) and time

period (DUL:hasEventDate). The property DUL:hasParticipant further links

them with the event visitors (foaf:Agent and subsequently with foaf:Person).

12 A live instance of the system is available online, with demonstrations for each of the

described use cases, via the following link: http://bit.ly/1ACmp6s


Mapping to a specific city is achieved through dbpedia-owl:City, further

linked with external GeoNames feature codes. To describe the nature of the

ingested data (static, semi-static, real-time), the Sampling primary class is

used and, respectively, its subclasses osmosys:StaticSampling,

osmosys:SemiStaticSampling, and osmosys:DynamicSampling. Following the

spatial and temporal annotations of the data inputs, the latter are further

specified according to their source (hasDataSource: Demographics,

ssn:SensorInput, SocialWebData). Classes and properties of the KR

framework’s ontology representing human-generated data sources were

particularly important in describing the social media streams for both example

cases. Especially, the classes osmosys:SocialMediaPlatform (containing

instances for Twitter and Instagram), osmosys:SocialWebData (subclass of

DUL:InformationObject), and osmosys:HumanSensor (subclass of ssn:Sensor

and equivalent to foaf:Person). The latter assisted in clearly defining whether

data streams referred to (isProducedBy) sensor devices (described by

ssn:SensingDevice) or social media inputs produced by people

(HumanSensor). In this way, a stakeholder can perform queries about, for

instance, the traffic flows on a specific day of the event, merging together

measurements (ssn:hasMeasurement) produced by traffic sensors and geo-

referenced tweets about the traffic conditions around the event site. In

addition, data inputs were enriched with thematic annotations, with regard to

the urban system they belong to. This is achieved via object properties, such

as refersTo, isAbout, isConstituentOf, isEmbeddedIn and others, which relate

data inputs to specific urban systems (e.g. Transport, CityEvent etc.).

Through the previously described mappings, RDF triples of the ingested

data were generated, enriched with spatial, temporal, and thematic

annotations. This allowed organizers and urban managers to perform

specialized queries, based on the issue at hand. The web-based KG further

enabled data exploration and visualization, facilitating urban analytics.


Fig. 3. Simplified view of the semantic data mapping into (parts of) the ontology

hierarchy for the two city-scale events (ALF and MDW)

6 Conclusions and Future Work

This paper introduced OSMoSys, a novel knowledge representation

framework for smart city planning and management that enables the semantic

integration of heterogeneous urban data from diverse sources. It further

showed how the framework’s three main components cater for a modular

system architecture that provides expansion and scalability potential, adapting

to the changing character of urban systems. OSMoSys enables the mutual

interaction among urban planners, city managers, decision-makers, and other

city stakeholders by providing tools that are capable of operating across the

disparate data silos.

In particular, the developed ontology tackles the semantic discrepancies

among urban data of different formats and speeds, by annotating them with

rich, machine-processable descriptions. The paper showed how data stemming

from smart devices and social media platforms could also be mapped to

concepts accommodated in the ontology hierarchy. In this way, these

increasingly important sources of information can easily be integrated in

urban analytics. Besides, the complementary web ontology browser and

interactive knowledge graph give stakeholders the opportunity to navigate the

ontology online and, further, perform data exploration and visualization.


Finally, the experimental implementation of OSMoSys in two real-world cases

as a back-end for an urban analytics system, demonstrated its potential in

semantically enriching and integrating historical (static and semi-static) and

real-time urban data from heterogeneous sources.

As part of future work, the plan is to conduct additional tests on mapping

open data (mainly municipal records and streams from mobile phones and

social media) from different sectors and various cities to the framework, so as

to further examine its validity and completeness. These tests will indicate the

potential necessity for incorporating new concepts in the ontology, or new

sub-ontologies. In addition, query support for linked open data requires

further development. Covering a wider spectrum of urban concepts, the

framework will constitute a valuable tool for urban planners and managers

and help them exploit the increasing amount of available data sources.

Acknowledgements. This research is funded by the Greek State Scholarships

Foundation I.K.Y. (by the resources of the Educational Program “Education

and Lifelong Learning”, the European Social Fund (ESF), and the EU

National Strategic Reference Framework (NSRF) of 2007-2013). It is further

funded by a scholarship of the Alexander S. Onassis Foundation. It is also

financially supported by the Foundation for Education and European Culture

(IPEP) and the A. G. Leventis Foundation.


References

Antoniou G, van Harmelen F (2009) A Semantic Web Primer. The MIT Press:

Cambridge, MA, USA

Bellini P, Nesi P, Rauch N (2014) Knowledge Base Construction Process for

Smart-City Services. In: Bilof R (ed) 19th International Conference on

Engineering of Complex Computer Systems (ICECCS 2014). IEEE, Los

Alamitos, CA, USA, pp 186–189

Bizer C, Heath T, Berners-Lee T (2009) Linked Data – The Story So Far.

International Journal on Semantic Web and Information Systems (IJSWIS)

5(3):1–22

British Standards Institution (BSI) (2014a) Smart Cities – Vocabulary. BSI

Standards, London

British Standards Institution (BSI) (2014b) Smart City Concept Model –

Guide to establishing a model for data interoperability. BSI Standards,

London

Compton M, et al (2012) The SSN Ontology of the W3C Semantic Sensor

Network Incubator Group. Web Semantics: Science, Services and Agents on

the World Wide Web 17:25–32

Domingue J, Fensel D, Hendler JA (eds) (2011) Handbook of Semantic Web

Technologies. Springer, Berlin Heidelberg

European Innovation Partnership on Smart Cities and Communities (EIP-

SCC) (2014) Operational Implementation Plan. EIP-SCC, Brussels


Euzenat J, Shvaiko P (2013) Ontology Matching (2nd ed). Springer, Berlin

Heidelberg

Falquet G, Métral C, Teller J, Tweed C (eds) (2011) Ontologies in Urban

Development Projects. Springer London, London

Gröger G, Kolbe TH, Nagel C, Häfele KH (eds) OGC City Geography

Markup Language (CityGML) Encoding Standard. Document Reference

Number OGC 12–019, version 2.0.0

Hu Y (2005) Efficient, High-Quality Force-Directed Graph Drawing.

Mathematica Journal 10(1): 37–71

Métral C, Falquet G, Cutting-Decelle AF (2009) Towards Semantically

Enriched 3D City Models: An Ontology-based Approach. In: Kolbe TH,

Zhang H, Zlatanova S (eds) GeoWeb Conference Academic Track–

Cityscapes. International Archives of Photogrammetry, Remote Sensing and

Spatial Information Sciences (ISPRS Archives) vol XXXVIII-3-4/C3

Vancouver, Canada, pp 40–45

Montenegro N, Gomes JC, Urbano P, Duarte JP (2012) A Land Use Planning

Ontology: LBCS. Future Internet 4(4):65–82. doi: 10.3390/fi4010065

Poveda-Villalón M, García-Castro R, Gómez-Pérez A (2014) Building an

Ontology Catalogue for Smart Cities. In: Mahdavi A, Martens B, Scherer R

(eds) eWork and eBusiness in Architecture, Engineering and Construction

(ECPPM 2014). CRC Press, Vienna, pp 829–836

Smart Cities Council (2014) Smart Cities Readiness Guides: The planning

manual for building tomorrow’s cities today. Smart Cities Council, Redmond,

WA, USA


Documents

Ontology-Based Data Integration from Heterogeneous …web.mit.edu/cron/project/CUPUM2015/proceedings/Content/... · ... age and gender distribution ... Linked Data [3] and Semantic